by George Michaleson
This article is looking at different measurements at the network level and at the end user level and examines the results seen by these measurements and some discussion of the variations between the two approaches.
Recent IPv6 statistics from the RIPE NCC show an accelerated uptake of IPv6 in Norway, both in terms of the number of allocated prefixes, and visible announcements in the routing system. This is based on a comparison over time of the amount of IPv6 addresses allocated to each economy, and the amount of visible prefixes per autonomous system (AS) in the routing tables each day.
Some have incorrectly interpreted this to mean that over 50% of the end-users in Norway have now access to IPv6. End-user IPv6 capability measurements being undertaken by APNIC suggests end-user access to IPv6 remains a current cause for concern in Norway, as it is in other economies. The graph below shows the percentage of IPv6 preference at the end-user level.
Note that this only includes data until mid May 2012. For the most up to date graph. If you click on the image, you will get to the most up to date graph and the underlying methodology.
Are these measures in conflict?
No, not really. One is a measure of capacity and capability in routing and forwarding, and the other is a measure of access network capability and end-host behaviours. There are many reasons why some routing active entities don’t show up in an end user measure: the AS may be servicing IPv6 content delivery while not offering IPv6 end user access services, or may be providing transit and data management services for others, and have no direct end user traffic.
Perhaps the AS is servicing segments of the user base who only gain access to the global Internet occasionally, or to restricted URLs, or not even the web but only VOIP (which we can’t measure in the APNIC technique.)
The difference in these two measurements not a conflict per se, but is exposing differences in what we see on the Internet, and different conclusions have to be drawn from each measurement technique.
APNIC’s measurement focuses on end user behaviour and the capability of end users to access a service using IPv6, and in large part, suggests that there is a continuing cause for concern with the very low levels of end use of IPv6, even when the end user’s Internet Service Provider in question may have associated IPv6 allocations visible in global routing.
The reasons for this very low level of uptake of Ipv6 by the Internet’s end user population are complex, and lie as much in the business dynamics of the ISP industry as much as in specific technology issues, but there is a widespread requirement to upgrade customer premises equipment (CPE) (Such as the DSL or cable modems used by residential broadband customers) to permit deployment of native IPv6 to and ISP’s customers. This particular topic has been covered in other articles such as the 2011 RIPE Labs article on the IPv6 CPE Survey by Marco Hogewoning and Mirjam KÃ¼hne.
In order to make an IPv6 prefix visible in the global routing system, a network engineer has to add several lines of text to their BGP definitions in a router configuration, and in order to prevent filtering by neighbors and up-streams, also should prepare a brief statement of their routing intentions to publicly list in an Internet Routing Registry (IRR). (this doesn’t directly alter how BGP works, but it does help prevent other ISPs from filtering out these route announcements). The total work is not great, and the impact on the core routing stability of the ISP is not great, so this is both easy to do, and a low-risk activity.
It should not be surprise to observe that many IPv6 address holders have in fact made an announcement, and if you look at the BGP routing tables, you will see a total of around 9,000 announced IPv6 prefixes, where the allocation records show a total of 10,000 address allocations, with some 5,521 Autonomous Systems announcing or carrying Ipv6 routes out of a total of 41,000 AS’s in use in today’s routing system (http://bgp.potaroo.net/v6/as2.0/). On this measure, 80% of allocation IPv6 address prefixes are visible. Of course, traffic engineering and the need to announce multiple sub-splits of an allocation alter this, but to a first approximation we can say that 80% visibility suggests ISPs who obtain an Ipv6 address block advertise it in to the routing system. This is good!
What else is needed on the provider’s side?
However, making a BGP announcement visible is by far the easiest part of deploying IPv6 in a network. The address resource has to be sub-structured, and assigned to subnets inside the internal routing system. Each internal router has to be configured with its own subspace, and likewise local support systems such as DNS, DHCPv6 and ACLs configured to make the subspace work in the local context: assign end-site addresses, map names to addresses, propagate internal routing.
This is not logistically different to IPv4, except that there are far fewer tools to manage this, and therefore far higher labour costs, and associated risk (in fact, where systems exist to manage network configuration in IPv4, adding IPv6 using a manual workflow is very probably next to impossible because the automatic systems will be overwriting configurations with IPv6 updates, replacing them with IPv4 only material).
What else is needed on the customer’s side?
There is the problem of getting the customer to actually request and configure IPv6 on their hosts. Now, by default, Vista, Windows 7, and OSX (along with Android and iOS) will in fact request IPv6 configuration information automatically. They will do this continuously once booted in the background, and are more than capable of working in an IPv6 environment out-of-the-box. Windows XP and other operating systems vary in their configuration, mostly being able to run IPv6 but rarely automatically configured to enable it (Linux and FreeBSD systems will detect and enable IPv6 but are not always configured to do this from installation).
The missing link is the CPE: the customer premises router, which attaches to the ISP and crucially mediates the link technology (cable, ADSL, modem, 3G) to provide the service to the users host. These devices typically run small embedded operating systems (often based on a stripped version of Linux) with reduced memory footprint and have few (if any) opportunities for field-upgrade of the code.
Although CPE have been known to need to be IPv6 capable for some time, the actual deployment rates of IPv6 capable units has been low: DocSys 3 modems, and some more recent ADSL2+ devices (including Fritz, Netcomm, Billion, Dynalink and D-link to name a few) are known to work in current deployments, but the overwhelming majority of CPE devices deployed today cannot and will not do IPv6.
APNIC’s measurement methodology is examining end-user IPv6 capability by use of fetches of ‘invisible’ one-pixel images, where each image is provided using different forms of protocol access: one image is available on both IPv4 and Ipv6 (Dual-stack), one on IPv4 only, one on IPv6 only, and one on an IPv6 literal address (which forces open automatic tunneling on some end user systems).
We use two different methods of delivering these tests to the end user:
- a flash based mechanism which requires no changes to a web page, and relies instead on the online advertising networks to deliver these tests to end users, using paid flash advertising to be displayed on the end user’s browser.
These mechanisms are generating around 1M individual end user measurements per day worldwide, and have generated a wealth of data for regions, economies and ISPs (http://labs.apnic.net/ipv6-measurement/).
The Situation in Norway
Analysis of the data for Norway suggests that the higher level of IPv6 deployment by network operators has not yet been matched by an increased uptake in IPv6 by Norwegian end users. There has been a surge of indirect use of IPv6, probably due to the ‘Opera’ Browser now including a speed-cache mechanism aimed at smart phone users on Android and iPhone, and this Opera capability uses a worldwide proxy deployment which is fully IPv6 enabled: so by indirect fetches, Norway shows a high number of users. But if you ask for direct access figures, the story relating to end user levels of use of IPv6 in Norway is more common with the rest of Europe.
At this time, Norway shows an IPv6 “capability” (including use of tunneling) of the order of 3.36% of the total user base (here “capability” is a forced measure intended to expose those users whose systems have an active IPv6 protocol stack, irrespective of whether it will be used in the normal course of using the Internet). If IPv6 “preference” (“preference” a stronger indication of access to IPv6 where the user will prefer to use the Ipv6 protocol when offerred the choice between IPv4 and IPv6 in a Dual-stack scenario) is measured, the figure is 0.35% of the total user base. These figures are essentially unchanged over the last 6 months, from a monthly sample size of over 1500 measurements (there is variance on a monthly basis, and the figure has been as high as 0.50% but its certainly not uniformly rising, nor into the high single digits yet).
The story is no better when examined by ASNs. Although a significant proportion of Norwegian ISPs demonstrate IPv6 routes in the global BGP, far fewer appear to be deployed in a way which customers can exploit. At this time, only 3 AS are consistently visible from Norway with end-user traffic:
More Norwegen networks are seen, but because the overwhelming use in Norway is via Opera-proxy, these measurements have been discounted in the APNIC measurement as they are in fact testing the capability of the Opera Proxy server set, rather than the capabilities of the end user’s environment.
APNIC has successfully measured the level of IPv6 access by 112 of the 175 ASNs in Norway but unfortunately most of these ASNs have generated insufficient data to be an accurate measure. If you only have 2 samples, and one of these samples showed a preference for using IPv6, then to declare that we have observed a 50% IPv6 preference would be stretching the data beyond conventional credibility! On the other hand, we are seeing the measurement tests being executed from IP addresses that are announced by these other 109 AS’s. This suggests that the APNIC measurement technique illustrates that 64% of the ASNs in Norway are ‘end user servicing’ ASes, and that whilst only a few prefixes inside each ASNs have been seen, we have a sense that 64% of the discrete routing entities service a population of end users. It is possible that this measure is an effective tool for distinguishing end-user from exchange/transit, content serving and other AS usage. This is a fruitful area for future study.
If we accept a significant drop in statistical validity and consider the top 50 Norwegian ASNs which have been seen in the experiment, and focus on the last 3 weeks of data, we see in fact there is some evidence of wider IPv6 availability to end users. The following 16 ASNs for instance, record significantly higher IPv6 preference values in this more loose measure:
|AS2119||TELENOR-NEXTEL Telenor Norge|
|AS8542||BKKB BKK Fiber|
|AS15765||MIMER Tafjord Mimer AS|
|AS41572||Hafslund Telekom Nettjenester|
There is therefore reason to hope that the IPv6 penetration in Norway is more than merely perfunctory, and is reaching some threshold of critical mass. However, it would be a mistake to believe the story is over: there is a long way to go in achieving a realistic worldwide deployment, but the signs are encouraging.
The Situation in Japan
As a comparison, Japan has been seen in the APNIC measurement in 43 ASNs, out of a total of 126 assigned to Japanese entities. The difference is quite marked: Japan has a far larger population (127 million) than Norway (5 million) and yet has broadly the same number of ASNs assigned to the economy. Clearly there is far more independently routed networks per capita in Norway than Japan, yet of the 43 seen ASNs in Japan 29 are providing statistically useful volumes of data.
The following 9 ASNs show any significant IPv6 deployment by the same loose count:
|AS4713||OCN NTT Communications Corp.|
|AS2510||INFOWEB Fujitsu Ltd.|
|AS17676||GIGAINFRA Softbank BB Corp.|
|AS10010||TOKAI Victokai Corp.|
|AS2497||IIJ Internet Initiative Japan Inc.|
APNIC compiles a regularly updated table of the relative address distribution by economy including basic GDP and population data from external sources, for both IPv4 and IPv6:
From this, we can see Norway ranks 29th in the world by % holdings in IPv4. By comparison, in IPv6 it stands at the 19th position. Once population, GDP, and AS visibility is factored in, the relative rankings of address holdings change radically. By announced IPv4 space, Norway ranks in 60th place, whereas by IPv6 it still ranks 19th. Therefore, considering the relative size, population and economic worth, Norway is substantially ahead of the field in relative terms, in IPv6 when address allocation data is used as the basis for national comparison. It’s therefore a matter of some concern that the measurements of the use of this address space, namely that of end user access remains at a low level. However, in context, it remains amongst the top 25 end-user access enabled economies which rank as follows, with Norway in 23rd place.
|Country||% of IPv6 at customer|
On APNIC Labs you can find a map showing the percentage of IPv6 end-user access by for each economy. You can also find more information about the methodology and other IPv6 measurements.
It is also notable that overall economic size or GDP is not necessarily a good indication of the likelihood of high IPv6 penetration. Smaller economies may be at an advantage in that a deployment involves less complexity, and these economies can higher penetration per capita compared to physically large, populous economies or ones with a highly distributed economic base.
Future work could usefully explore the relationship between the amount of Internet resources deployed in economies and regions, the amount seen, and the relativities between these distributions. It would help explain systematic biases in the measurements, and also inform questions of the GDP and economic relativities of the costs and benefits of the IPv6 network deployment.